Skip to content
X · @teortaxesTex · X / Twitter

Infra vs architecture is partially a matter of confidence. You can hedge on design and then throw workhours at optimizing the hell out of your serving…

Infra vs architecture is partially a matter of confidence. You can hedge on design and then throw workhours at optimizing the hell out of your serving system. Or you can do everything right from the start, and find a better regime.Tremendous effort. If only V4 didn't exist.Xiuyu Li: Hybrid SWA only saves inference cost